mrpositron / paper2tex

Extracting LaTeX equations from PDF

Paper2Tex (discontinued)

Check Nougat: Neural Optical Understanding for Academic Documents

The following project is a tool to extract equations from the research papers (images, PDFs, etc.) and convert it into latex code.

This project is heavily utilizes the following projects:

• ScanSSD: Scanning Single Shot Detector for Math in Document Images
• ScanSSD (fork)
• pix2tex - LaTeX OCR

Credit goes to the authors of the above projects, @MaliParag, @lukas-blecher, @jjdredd.

How to use?

paper2tex.ipynb is the main notebook. It contains the code to extract equations from the paper. The notebook is self explanatory.

Example

Extracted equations are in boxes with yellow border. In top left corner of each box, there is a number which is the id of the equation, and on the top right corner.
The extracted equations are:

• $$\text{id:}0 \Rightarrow {\frac{1}{N}}\sum_{i=1}^{N}\ell(\mathbf{x}_{i},\Theta)$$
• $$\text{id:}1 \Rightarrow \Theta_{2}\leftarrow\Theta_{2}-\frac{\alpha}{m}\sum_{i=1}^{m}\frac{\partial F_{2}({\bf x}_{i},\Theta_{2})}{\partial\Theta_{2}}$$
• $$\text{id:}2 \Rightarrow \ell=F_{2}(F_{1}(\mathbf{u},\Theta_{1}),\Theta_{2})$$
• $$\text{id:}3 \Rightarrow {\frac{1}{m}}{\frac{\partial\ell(\mathbf{x}_{i},\Phi)}{\partial\Theta}}$$
• $$\text{id:}4 \Rightarrow \ell=F_{2}(\cdot)$$

Things to do

• Add a notebook to extract equations from the paper.
• Implement a GPU version of the code.
• Upload it to the colab
• Find a way to use inference LaTeX-OCR in batch mode.
• Detect paper borders